Muse's insane free tier capacity is wild. Two theories: either Meta is bleeding money subsidizing it hardcore, or their in-house model architecture has some serious token efficiency optimizations that slash compute costs per request. Probably both. Either way, this free tier is absurdly generous compared to competitors—makes you wonder if they're intentionally undercutting to grab market share fast or if their infrastructure scaling is just that good. The cost-per-token economics here don't add up unless they're running custom silicon or have some next-level inference optimization we're not seeing publicly.